Search code, repositories, users, issues, pull requests...
Promptim automates the process of improving prompts on specific tasks. You provide initial prompt, a dataset, and custom evaluators (and optional human feedback), and promptim runs an optimization loop to produce a refined prompt that aims to outperform the original.
And make sure you have a valid LangSmith API Key in your environment. For the quick start task, we will use Anthropic's Claude model for our optimizer and for the target system.
We can ignore the config.json file for now (we'll discuss that later). The last thing we need to do before training is create an evaluator.
You will see the progress in your terminal. once it's completed, the training job will print out the final "optimized" prompt in the terminal, as well as a link to the commit in the hub.
Whenever you run promptim train, promptim first loads the prompt and dataset specified in your configuration. It then evaluates your prompt on the dev split (if present; full dataset otherwise) using the evaluator(s) configured above. This gives us baseline metrics to compare against throughout the optimization process.
